JPH09305469A - File exclusive control system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a file exclusive control system which advances the execution of jobs while evading a scramble for a block with other jobs when a job which cames to a deadlock restarts its process. SOLUTION: This system is provided with a lock management means 12 which registers a block in a block number list 21 and decides whether the block can be locked or comes to a deadlock, an image gathering means 13 which gathers block contents before update, a check point file 31 which stores the block contents before update and a program at check point time, a check point means 15 which makes the block list empty at the check point time and stores the program contents, and a rollback means 16 which writes the data file and program contents back at deadlock time. When the job restarts accessing the data file from the last check point after coming to the deadlock, block groups up to the deadlock occurrence time registered in the block number list are locked collectively at the start.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a file exclusive control system in which a plurality of jobs can simultaneously access the same data file without any contradiction, and in particular, it is possible to promptly restart processing after a deadlock. Related to the exclusive control method of the file.

[0002]

2. Description of the Related Art Conventionally, as an exclusive control system of this kind of file, there is one disclosed in, for example, Japanese Patent Laid-Open No. 24-48648. The exclusive control system disclosed in Japanese Unexamined Patent Publication No. 3-24648 aims to preserve the file and shorten the processing time for deadlocks and other failures, and has a simultaneous access control means, a large capacity buffer management means, and a post-update function. The image collecting unit, the checkpoint unit, the updated image saving unit, the rollback unit, and the recovery unit are included.

The simultaneous access control means executes exclusive control for the job and detects the occurrence of deadlock, and the large capacity buffer management means reads the referenced / updated block from the data file into the main storage area and updates it. The pre-update image buffer and the post-update image buffer are retained for the block. The post-update image collecting means stores the post-update image given from the job in the post-update image storage file allocated for each job.

When a checkpoint is requested, the checkpoint means writes the contents of the post-update image buffer in the main storage area to a data file, discards the pre-update image buffer, and describes the space of the post-update image storage file. Register the child as part of the post-update common image storage file in the system and separate it from the user program.

The post-update image saving means saves the contents of the common post-update image storage file in the system to the post-update image save tape when the amount of the common post-update image storage file in the system exceeds a certain amount, and saves the file space. To release.

When a failure such as a deadlock occurs, the rollback means writes the contents of the pre-update image buffer in the main storage area to the post-update image buffer, deletes the update flag, and empties the post-update image storage file. . In addition, the recovery means recovers the contents of the data file from the contents of the post-update image storage file for the job that ended abnormally during checkpoint processing, and
Complete the checkpoint process.

[0007]

In the conventional file exclusive control system described above, since all jobs are treated equally, there is a problem that execution of a job in which a deadlock occurs may be delayed. there were. The reason is that even if a deadlock occurs and the job starts over from the previous checkpoint, and it is obvious that the job execution will be delayed, it is simply one of the jobs that simultaneously accesses the file. Treated,
This is because it does not provide a means for prioritizing the job.

It is an object of the present invention to, when a job in which a deadlock has occurred restart processing, collectively lock the block that was locked and the block that was attempted to be locked by the time the deadlock occurred. The object of the present invention is to provide a file exclusive control system capable of advancing the execution of a job while avoiding contention for a block with another job with respect to access up to the time when the error occurs.

[0009]

To achieve the above object, the present invention divides a data file into blocks of a certain size, and a plurality of jobs simultaneously copy the same data file by exclusive control in block units. In a file exclusive control system that can be accessed without contradiction, when the data file is accessed or the checkpoint is acquired by the job, an attempt is made to access between the file update means for operating the necessary control means and the checkpoint. A block number list for registering the block number, and a lock management unit for registering the block to be accessed in the block number list and determining whether the block can be locked or deadlock occurs,
When the data file is updated, an image collection unit that collects the block contents before the update prior to the update, the block contents before the update collected by the image collection unit, and the program contents at the time when a checkpoint is taken A checkpoint file for storing the data, and a file input / output means for reading / writing the data file,
From the checkpoint means that saves the program content by emptying the block numbers registered in the block number list at checkpoint, and the rollback means that writes back the data file and program content from the checkpoint file when a deadlock occurs. Composed,
When a job resumes access to a data file from the previous checkpoint after a deadlock occurs, the deadlock is achieved by first locking all the blocks registered in the block number list up to the deadlock occurrence time. The feature is that the execution of the job that falls into the above and the access of the data file can be prioritized.

According to the present invention of claim 2, the lock management means judges whether or not it is the first file access immediately after the occurrence of the deadlock, and if it is the first file access, the block group registered in the block number list. It is characterized by locking all at once.

According to the present invention of claim 3, when the deadlock is detected, the lock management means writes the information indicating the occurrence of the deadlock in the block number list and operates the rollback means. It is characterized by

According to the present invention of claim 4, the lock management means judges whether or not it is the first file access immediately after the occurrence of the deadlock, and if it is the first file access, the registered block group is selected. Lock them all together,
Further, it is characterized in that the information indicating the deadlock occurrence described in the block number list is erased.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a first embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a file exclusive control system according to an embodiment of the present invention.

Referring to FIG. 1, the file exclusive control system includes a data processing device 10 for accessing a file by program control, a main memory device 20 made of a semiconductor, and an auxiliary device made of a disk device or the like. The storage device 30 includes a job group 40 including a plurality of jobs 41 to 43, and a data file 50 including a plurality of blocks 51 to 55.

When the job accesses the data file 50 or takes a checkpoint, the data processing apparatus 10 can lock the file updating means 11 for activating necessary lower means and the block to be accessed. Lock management means 12 for determining whether or not a deadlock occurs, before-image collection means 13 for collecting the block contents before update when the data file 50 is updated, and the data file 5.
A file I / O means 14 for inputting and outputting to and from 0,
A checkpoint unit 15 for saving the program contents by emptying the block numbers registered in the block number list at the checkpoint, and a rollback unit 16 for writing back the data file and the program from the checkpoint file when a deadlock occurs. .

The main storage device 20 is provided with a block number list 21 for registering the block numbers to be accessed during the checkpoint.

The auxiliary storage device 30 includes a checkpoint file 31 for storing the collected block contents before update and the program contents at the time when the checkpoint is acquired. The checkpoint file 31
Is preferably prepared for each job, but in the present embodiment, the configuration is commonly used for each job in order to simplify the description.

When a job accesses the data file 50, the file updating means 11 first activates the lock managing means 12. The lock management unit 12 registers the numbers of the blocks 51 to 55 of the data file 50 to be accessed in the block number list 21 and determines whether the block can be locked or deadlock occurs with another job. Or not.

When the block can be locked, the file updating means 11 operates the file I / O means 14 or the before image collecting means 13 in response to a read or write request of the data file by the job.

Before-image collecting means 13 stores the contents of the block in the checkpoint file 31 before the block of the data file 50 is written. After that, the file updating means 11 operates the file I / O means 14 as in the case of the read request. The file I / O unit 14 executes the reading process or the writing process of the contents of the block according to whether the job request is reading or writing.

When the job takes a checkpoint, the file updating means 11 operates the checkpointing means 15. The checkpoint means 15 deletes the contents of the block group stored in the checkpoint file 31 and stores the program image of the job in the checkpoint file 31. At the same time, the block numbers registered in the block number list 21 are deleted.

When the deadlock is detected, the lock management means 12 writes information (flag or the like) indicating that the deadlock has occurred in the block number list 21 and operates the rollback means 16. Rollback means 1
Reference numeral 6 writes back the program image of the job and the block of the data file 50 from the checkpoint file 31 and enables re-execution from the immediately preceding checkpoint.

A feature of the first embodiment of the present invention is that the lock management means 12 executes the following control. That is, it is determined whether or not it is the first file access immediately after the occurrence of the deadlock, and if so, the blocks registered in the block number list 21 are collectively locked.

When the block number list 21 contains information indicating that a deadlock has occurred, the lock management means 12 locks the registered block groups collectively,
The information on the occurrence of deadlock in the block number list 21 is deleted.

Next, referring to FIGS. 1 and 2 to 7,
The operation of the first embodiment of the present invention will be described.

A request for a job to access the data file 50 or take a checkpoint is received by the file updating means 11. As shown in FIG. 2, the file updating means 11 determines the content of this request (step 201), and if it is a file access request, operates the lock management means 12 and otherwise operates the checkpoint means 15 (step 202 or Step 207).

As shown in FIG. 3, the lock management means 12
First, it is determined whether or not the information on the occurrence of deadlock is written in the block number list 21 (step 30).
1). If not, then it is determined whether the block number to be accessed this time is already registered in the block number list 21 (step 302).

If the deadlock information is not written and the block number is registered, the process immediately returns to the control of the file updating means 11.

If the deadlock information is not written and the block number is not registered, the block number is registered in the block number list 21 (step 30
3) It is determined whether the block can be locked (step 304), and whether deadlock occurs (step 306).

If the lock is possible, the block is locked (step 305) and the process returns to the file updating means 11. If the block cannot be locked but deadlock does not occur, it waits for another job to release the block (step 307) and then determines whether it can be locked again (step 304).

If the block cannot be locked and a deadlock occurs when waiting for the release of the block, information on the occurrence of the deadlock is written in the block number list 21 (step 308) and the rollback means 16 is operated (step 308). Step 309). After returning from the rollback means 16, the lock management means 12 notifies that the deadlock has occurred and returns to the file update means 11.

As shown in FIG. 2, the file updating means 11
Then, if there is no notification of the occurrence of deadlock, it is determined whether the file access type is read or write (steps 203 and 204), and the read agate file I / O means 14 is operated (step 206). If writing, first the before image collecting means 13 is operated, and then the file I / O means 14 is operated (step 20).
5, step 206), satisfy the job request and return to the job.

As shown in FIG. 4, since the before image collecting means 13 stores the block image before updating, the block still has the checkpoint file 31.
If not, the block to be updated is read from the data file 50 and written to the checkpoint file 31 (step 401 to step 403).

Further, as shown in FIG. 5, file I / O
The means 14 reads the target block from the data file 50 or writes the target block to the data file 50 depending on whether the file access type is read or write (step 501 to step 50).
3).

If the request from the job is a checkpoint in step 201 of FIG. 2, the file updating means 11 operates the checkpoint means 15 (step 207).

Here, as shown in FIG. 6, the checkpoint means 15 deletes the block numbers registered in the block number list 21 and the block image before update written in the checkpoint file 31 ( Steps 601 and 602), the program image is saved in the checkpoint file 31 (step 6).
03). After that, the process returns to the file updating means 11 and returns to the job from there.

If the lock management means 12 determines that a deadlock will occur, the rollback means 16 checks the data file 50 and the job in order to restore them to the state of the previous checkpoint, as shown in FIG. All the blocks written in the file 31 are written back to the data file 50 (steps 701 to 704), and the program image is written back (steps 705, 706). After this, lock management means 1
2 notifies the file updating means 11 of the occurrence of deadlock.

The file updating means 11 immediately returns to the job when the occurrence of the deadlock is notified. The program image of the job has been restored by the rollback means 16 at the time of the immediately preceding checkpoint, and the job will resume processing from there, and the data file 50 is accessed in the same order as before.

As shown in FIG. 3, in the lock management means 12, when the information on the occurrence of deadlock is written in the block number list 21, all the block numbers registered in the block number list 21 are collectively shown. At the same time as locking (steps 310 to 312), the information on the occurrence of deadlock in the block number list 21 is erased (step 313).

Since the file exclusive control system according to the embodiment of the present invention operates as described above, when a deadlock occurs and the processing is restarted from the immediately preceding checkpoint, the deadlock is performed at the first file access. Since the blocks that have been accessed and the blocks that have tried to access are collectively locked by the time when occurs, normally, the locking process that occurs each time a different block is accessed becomes unnecessary. Therefore, the deadlock with another job does not occur again before the deadlock occurs last time, and the process can be promptly recovered and further advanced.

Next, a specific operation example of the embodiment of the present invention will be described in detail. In FIG. 8 and FIG. 9, the job 41 and the job 42 of the job group 40 are shown in block 5 respectively.
The contents of the block number list 21 and the checkpoint file 31 when the data file 50 composed of five blocks 1 to 55 are being accessed are shown.

Here, the job 41 reads the block 51 and writes the block 52. Further, the writing to the block 53 is awaited because the job 42 is reading the block 53. The job 42 reads the block 53 and writes the block 54. Further, the job 42 has acquired a checkpoint before this.

From the state shown in FIGS. 8 and 9, when the job 42 tries to write out the block 55, the file updating means 11
First operates the lock management means 12 (steps 201 and 202 in FIG. 2). The lock management means 12 judges from the block number list 21 in the state of FIG. 8 (steps 301 and 302 in FIG. 3), registers the block 55 in the block number list 21, and locks it (step 30 in FIG. 3).
4, 305).

Subsequently, the file updating means 11 operates the before image collecting means 13 (step 2 in FIG. 2).
03-205). Before image collecting means 13
The block 55 is read from the data file 50 and written to the checkpoint file 31 (steps 401 to 403 in FIG. 4). Finally, the file updating means 11
Causes the file I / O means 14 to operate (step 206 in FIG. 2), and the file I / O means 14 causes the block 55
Is written in the data file 50 (step 502 in FIG. 5), and then the process returns to the job 42.

When the job 42 takes a checkpoint from the states shown in FIGS. 8 and 9, the file updating means 11 operates the checkpoint means 15 (step 2 in FIG. 2).
07). The checkpoint unit 15 deletes the blocks 53 and 54 registered for the job 42 from the block number list 21 (step 601 in FIG. 6).

Further, since the job 42 has written the block 54, the checkpoint means 15 deletes the contents of the block 54 before the update from the checkpoint file 31 (step 602 in FIG. 6). Finally, the checkpoint means 15 writes the program image of the job 42 to the checkpoint file 31 (see FIG. 6).
Step 603), and the process returns to the job 42.

It is assumed that the job 42 tries to write out the block 52 from the states shown in FIGS. In this case, the lock management means 12 registers the block 52 in the block number list 21 (step 303 in FIG. 3).
After this, the lock management means 12 uses the block number list 2
From FIG. 1, the job 41 has already locked the block 52, but the job 41 has already been made to wait by the job 42, so a deadlock is detected (step 3 in FIG. 3).
04, 306), the deadlock information is written in the block number list 21, and the rollback means 16 is operated (steps 308, 309 in FIG. 3).

The roll-back means 16 first executes the job 42.
All the contents of the updated block are written back from the checkpoint file 31 to the data file 50 (steps 401 to 404 in FIG. 4). In FIGS. 8 and 9, the block 54 is written back. Next, the rollback means 16
Is the state when the job 42 took the checkpoint last time from the checkpoint file 31.
It writes back itself (steps 405 and 406 in FIG. 4). This causes the job 42 to resume processing from the time of the previous checkpoint.

When the deadlock is returned to the job 42, the block 53 is released. As a result, the waiting job 41 can also continue processing.
Here, it is assumed that the job 41 ends by writing the block 53, and the new job 43 is the job 41.
Let us begin to update the blocks of the data file 50 in the same order as.

When the job 42 resumes processing, it first tries to read the block 53. In this case, the lock management means 12 confirms the information on the deadlock occurrence described in the block number list 21, locks the blocks registered therein collectively, and finally obtains the information on the deadlock occurrence. Erase (step 301, step 3 in FIG. 3)
10 to 313). In this example, the block 53, the block 54, and the block 52 in which the deadlock has previously occurred are locked together.

On the other hand, the job 43 attempts to write out the block 52 after reading the block 51. However, since the job 42 is already locked in the block 52, the job 3 is kept waiting and the job 42 is processed. Can continue. As a result, the job 42 can proceed with the writing of the block 54 and the writing of the block 52, with the job 43 kept waiting.

Although the present invention has been described with reference to the preferred embodiments, the present invention is not necessarily limited to the above embodiments.

[0053]

As described above, according to the present invention, at the time of the first file access after the deadlock, the blocks up to the time when the bad lock occurs are collectively locked, so the job for which the deadlock is returned is returned. However, when the process is restarted from the immediately preceding checkpoint, the file access is prioritized and the process can be restarted promptly.

That is, when a deadlock occurs and the processing is restarted from the immediately preceding checkpoint, the blocks accessed up to the time when the deadlock occurs at the time of the first file access are collectively locked Therefore, the lock process that occurs each time a different block is accessed becomes unnecessary. Therefore, the deadlock with another job does not occur again before the deadlock occurs last time, and the process can be promptly recovered and further advanced.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a file exclusive control system according to an embodiment of the present invention.

FIG. 2 is a flowchart for explaining an operation of a file updating unit.

FIG. 3 is a flowchart for explaining the operation of the lock management means.

FIG. 4 is a flowchart for explaining the operation of the before image collecting means.

FIG. 5 is a flowchart for explaining the operation of the file I / O means.

FIG. 6 is a flowchart for explaining the operation of the checkpoint means.

FIG. 7 is a flowchart for explaining the operation of the rollback means.

FIG. 8 is a diagram showing an example of contents of a block number list.

FIG. 9 is a diagram showing an example of the contents of a checkpoint file.

[Explanation of symbols]

 10 data processing device 11 file updating means 12 lock managing means 13 before image collecting means 14 file I / O means 15 checkpoint means 16 rollback means 20 main memory 21 block number list 30 auxiliary memory 31 checkpoint file 40 job group 41 to 43 jobs 50 data files 51 to 55 blocks

Claims (4)

[Claims] 1. A file exclusive control system in which a data file is divided into blocks of a certain size, and a plurality of jobs can simultaneously access the same data file without contradiction by exclusive control in block units, When a data file is accessed or a checkpoint is acquired by a job, a file updating means that operates the necessary control means, a block number list that registers the block number that was attempted to be accessed during the checkpoint, and an attempt to access Lock management means for registering a block in the block number list to determine whether the block can be locked or deadlock occurs, and when the data file is updated, the block contents before the update are updated before the update. Image collecting means for collecting, and the image A checkpoint file for storing the block contents before update collected in the collecting means and the program contents at the time when the checkpoint is taken, a file input / output means for reading and writing the data file, and a block number list for the checkpoint It consists of a checkpoint means that saves the program contents by emptying the registered block number, and a rollback means that writes back the data file and the program contents from the checkpoint file when a deadlock occurs. When the job resumes access to the data file from the last checkpoint, the block group registered up to the deadlock point registered in the block number list was locked at the beginning, resulting in a deadlock. Exclusive control system files, characterized in that the access of job execution and data files to allow priority. 2. The lock management means determines whether or not it is the first file access immediately after the occurrence of the deadlock, and if it is the first file access, locks the block groups registered in the block number list all together. The exclusive control system for a file according to claim 1, wherein the exclusive control system is for a file. 3. The lock management means, when detecting a deadlock, writes information indicating the occurrence of the deadlock in the block number list and operates the rollback means. File exclusive control system described in. 4. The lock management means determines whether or not it is the first file access immediately after the occurrence of the deadlock, and if it is the first file access, locks the registered block group at a time, and further, the block number. 4. The file exclusive control system according to claim 3, wherein information indicating occurrence of deadlock in the list is deleted.